NVIDIA / NVIDIA/NemoClaw

feat(ci): publish review queue evidence for read-only consumers

Open
#11,489 1 comment 0 reactions 0 assignees View on GitHub
Dominant language
TypeScript
Stars
22.5k
Forks
3.1k
Avg merge
1d 1h
Merged PRs (30d)
715

Description

## Decision

Accept. Record the maintainer-approved cross-repository implementation requested for the GitHub Monitor Review Queue.

Accountable maintainer: @jyaunches. Placement: existing Advisor and E2E producer tooling and owning repository guidance. Consumer: the GitHub Monitor Review Queue.

## Outcome

Publish versioned, validated, read-only evidence for the latest PR commit: Advisor completion and blocker count; complete E2E recommendations; and E2E execution identity and outcomes. Missing or stale evidence must remain unknown.

## Scope and compatibility

- Reuse the structured finding ledger proposed by #11047; coordinate its ownership before changing the specialist output contract.
- Preserve the deterministic risk-plan floor separately from additional Advisor recommendations. Preserve required and optional classifications; the consumer requires every recommendation to pass.
- Require an explicit no-tests-needed decision. Empty workflow dispatch selectors select the default suite and cannot represent no tests.
- Bind evidence to repository, PR, candidate/base/workflow commits, workflow run and attempt. Require every expanded execution for fan-out selectors.
- Reuse trusted E2E dispatch receipts and reconciliation. Document supported arguments and ambiguous dispatch handling without duplicate runs.
- Preserve sandbox credential separation, publisher isolation, workflow authorization, and hardware opt-ins. This work does not authorize merges, live tests, automated repairs, or changes to merge gates.

## Lifecycle and validation

The producer owns schema versions and fixtures. Consumers reject unsupported, missing, expired, stale, malformed, or incomplete evidence. Rollout begins when the producer workflow changes reach trusted main; earlier runs remain unknown.

Use focused deterministic tests for schema validation, explicit empty results, optional recommendations, selector expansion, run/attempt identity, and stale or missing evidence. Validate workflow trust boundaries and the required PR checks. Live E2E is not needed to test receipt parsing.

## Existing evidence

Canonical main inspected at fd98dc2aeb3a37579efbf005a0e2195f7bf20a03. Advisor completion currently publishes a run link. Risk-plan version 22 and nemoclaw-e2e-dispatch-v2 already exist. PR #11047 proposes a specialist finding ledger and overlaps the Advisor output owner.

Contributor guide

Open the contributing guide

Research direction

Start with PR #11047 and the existing Advisor and E2E producer tooling, including current Advisor run-link publication, risk-plan version 22, and nemoclaw-e2e-dispatch-v2. Run the focused deterministic tests for schema validation, selector expansion, run/attempt identity, and stale or missing evidence. Done means versioned, validated read-only evidence is published and consumers can reject incomplete or stale results without changing merge gates.

Written by the indexing model from the issue text.

Assessment

Tech stack
github-actions, typescript
Domain
ci-cd, devops, testing
Issue type
Feature
Difficulty
5/5
Estimated time
Over a week
Activity status
Active
Clarity
Mostly clear
Newbie friendliness
35/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.